HoneyDew¶

Your shared task board with your AI assistant.
HoneyDew is a live Kanban workspace where you and your AI agent collaborate on real work. Assign tasks, track progress, review outputs, and close the loop — together. It turns a one-off chat with your AI into an ongoing working relationship.
Built for OpenClaw users. Works with any agent that can call a REST API.
Why HoneyDew¶
Most AI interactions are ephemeral — you ask, it answers, the context vanishes. HoneyDew gives your agent a persistent workspace alongside you, so nothing gets lost and nothing goes unfinished.
- A shared source of truth — Both you and your agent read and write the same board. No more copy-pasting outputs into a tracker you maintain alone.
- Real handoffs — Assign a card to your agent, let it work, then have it hand the result back to you — with full history on the card.
- Work while you sleep — With a simple board check on each heartbeat, your agent works on its todo list while you're out of office.
- Works the way you think — Kanban columns, priority levels, due dates, labels, and multi-board support. A task board your agent actually uses.
- Zero friction — Two shell commands, no cloud account, no API key. Runs entirely on localhost.
- Open API, open tools — A REST API, a Python tools module, and an OpenClaw skill. Wire it up however you work.
How It Works¶
You create a task ─────► Agent picks it up ─────► Agent delivers results ─────► You review & close
(To Do) (In Progress) (Done / Blocked) (iterate or done)
- You assign a task. Create a card — "Refactor the auth module, add tests" — and assign it to your agent.
- Your agent picks it up. It reads the board, moves the card to In Progress, and gets to work.
- It posts its output. The agent updates the card with results and moves it to Done (or flags it as Blocked if it needs help).
- You review and iterate. See the result on the card, leave a note, reassign, or close it out.
Your agent can also check for new tasks independently — instruct it to poll HoneyDew for incomplete work and step away.
Getting Started¶
1. Clone¶
2. Install¶
This installs backend and frontend dependencies and creates a config.json from the example. The installer will prompt you for your name and your agent's name.
3. Start¶
Both servers launch together:
| Service | URL |
|---|---|
| Board UI | http://localhost:5173 |
| API | http://localhost:8000 |
| API Docs | http://localhost:8000/docs |
4. Connect your agent¶
Choose the path that fits your setup:
| Path | Best for |
|---|---|
| OpenClaw skill | OpenClaw users — one command, fully automatic |
| Share SKILL.md | Any agent — drop the context file into your system prompt |
| Python tools | Scripted or structured integrations |
What's Next¶
- Configuration — Customize profiles, board names, and columns.
- API Reference — Every endpoint, request body, and query parameter.
- Agent Integration — Three ways to connect your agent.
- For Agents — Condensed reference your agent can use directly.
- Python Tools — Full
KanbanToolsmethod reference. - Security — Deployment and trust model.
Tech Stack¶
| Layer | Technologies |
|---|---|
| Backend | FastAPI, SQLAlchemy 2.0, Pydantic v2, SQLite |
| Frontend | React 18, TypeScript, Vite, Tailwind CSS, @dnd-kit |
| Agent Tools | Python, httpx |